In an adding experiment when a time limit is used and the units in the task are as large as those employed here, that is, columns of 10 digits each, in which each column is added and the sum recorded separately, it must inevitably happen that when the signal to stop is given, some subjects will have a column only partly added. Obviously they can not be allowed to complete the column. The possibility of having the subject draw a mark through the column and write down the sum to that point, was considered. It seemed, however, that this would introduce the possibility of much irregularity and that as the time limit allowed for the task was 10 full minutes without interruption, it would be better to sacrifice this fraction of achievement in favor of a more simple and concise method of procedure. On the average, over 50 columns were correctly added during the 10-minute interval. The average loss could thus be not more than 2 per cent.

In the oft-repeated instructions, which always preceded the addition experiment, the matter of accuracy was given much more prominence than speed. The men were made to understand definitely that a column added incorrectly had better not have been added at all, so far as the score of the subject was concerned. In tables 155 and 156, which are of the usual form, the data presented give first place to accurate work. Two values are entered for each subject: (1) the number of columns of 10 digits correctly added during the 10-minute interval, and (2) the percentage of the total number of added columns whose sums were incorrect. For illustration, Bro of Squad A on September 29 added a total of 50 columns. Of these, 8 were incorrectly added. He therefore has a score on that date, as shown in table 155, of 42 columns correct and 16 per cent of errors. Can on the same date added a total of 30 columns, of which 6 were in error. His score is consequently 24 and 20 for correct columns and per cent error, respectively. To convert the errors into per cent makes possible the more direct comparison between subjects. The total number of columns added, if desired, may be computed from the two values in each of tables 155 and 156. For example, if the score for columns correct is 42 and the per cent of errors is 16, then 42 is 84 per cent of the total columns.

The individual variations among the members of Squad A are shown in the low-diet averages in table 155. Fre, Tom, and Pea average the largest number of correct columns in the time limit, with the values of 76.0, 61.8, and 58.3. Subject Fre was a remarkably good adder, but his case is of little interest to us, since he did not remain in the squad and the data are very few. Of the other men, the two previously mentioned (Tom and Pea) were at the head of the group in the normal experiment on September 29. In small percentage of errors these men do not lead, however, in the average during the low-diet period. Bro and Gar, with 4.3 and 8.0 per cent, have somewhat the advantage. Bro on October 13 and Pea on January 26 showed the only instances in which the subject's entire performance was without error. While there is considerable fluctuation in an indi-vidual's record from experiment to experiment, yet in general this is not such as to make it apparent that one individual influences predominantly the average for any one date. Five of the subjects whose records enter into the final averages for the squad, that is, Can, Gul, Moy, Pec, and Vea, have average low-diet errors of 19 per cent or above. Not one of these 5 subjects on any of the 9 dates which fell within the reduction period has a percentage of error less than 10 per cent. The largest is for Pec on October 13, viz, 40.7 per cent. This is somewhat compensated in the average by the fact that Moy on the same date had 26.1 per cent, which was a little below the average for him during the first part of the experiment. The errors for Gul were large, but during the greater part of the experiment they were quite consistent. Beginning with September 29 they are, in order, 25.0, 22.4, 24.5, 25.0, 18.3, 21.0, 23.8, 21.2, 12.3, and 12.7 per cent. The average score for Squad A during the whole of the reduction period, as shown at the bottom of the right-hand column in table 155, is 45.1 columns correct, with 15.5 per cent of errors.

Table 155. - Squad A - Results In 10-Minute Addition Test

Table 156. - Squad B - Results In 10-Minute Addition Test

In the case of Squad B the normal average performance shows that the individuals who made up this squad demonstrated a larger average variability than was found in Squad A. The scores for columns correctly added range from 85.8 and 68.8 for Tho and How, respectively, to 14.8 and 9.0 for McM and Sch. The data for the last two subjects are only fragmentary, as was also the case with Spe of Squad A. Nevertheless, there is somewhat more variability in Squad B, since no one in Squad A did so well as Tho and How of the former, and moreover, the scores of Har and Ham (22.6 and 23.0), both of Squad B, are slightly smaller than the smallest score with Squad A, i. e., Can, 25.3.1 The average normal performance for Squad B is 41.9 correct columns - about 3 columns less than the low-diet average of Squad A.

In the matter of percentage of errors, with the exception of Sch and McM, whose data, particularly those of the former, have been mentioned as incomplete and do not enter into the average for Squad B, the values are in general in the same range as those found for Squad A. The final normal average of 11.8 per cent for Squad B is 3.7 per cent less than that found for Squad A. It is noteworthy that the two subjects of Squad B, who lead with the largest number of columns correctly added, are also in the lead in the smallest percentage of errors of any of the 10 subjects from whom the final squad average is drawn. In comparing Squad B's normal average with the average for the three reduced-diet dates, only minor fluctuations are noted. Seven of the 10 subjects show increase in the number of columns correctly added. Only 4 showed a reduction in the percentage of errors. The final average for this period shows an increase of 2 columns in the 10 minutes, i. e., 43.9 compared to 41.9, and a decrease of 0.7 per cent in errors, 11.8 to 11.1 per cent.